three essays on the cutting edge of credit spread modeling · 2014. 12. 5. · 1 accurate pricing...
Post on 25-Feb-2021
2 Views
Preview:
TRANSCRIPT
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Three essays on the cutting edge of credit spreadmodeling
Alberto Fernandez Munoz de MoralesAdvisor: Alfonso Novales
UCM-QFB
December 16, 2014
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Outline
1 Credit spread modeling effects on counterparty risk valuationadjustments: a Spanish case study
IntroductionThe modelThe calibrationCVA-DVA calculations
2 Model risk in credit valuation adjustments: a Spanish case studyIntroductionDefault correlationSpread modeling
3 Credit spread modeling from a regulatory perspectiveCalculating VaR for CVABacktesting a risk factor under IMM
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Motivation
Question
Why would a financial entity need to model the behavior ofcredit spreads?
Three main reasons:1 Accurate pricing of derivatives (CVA-DVA adjustments).2 Parametric CVA VaR calculation.3 Risk factor modeling under IMM approach.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
CVA-DVA
Recent accounting standards demand adjustments on theprice of a derivative.
If in a given transaction the counterparty of a bank defaultedand the PV at default was positive for the bank, it would onlybe able to get a recovery fraction.
However, if such PV was negative, the bank would have topay in full.
Such asymmetry translated into a Credit ValuationAdjustment term (CVA).
This term needed to be subtracted from the net present valueof the equivalent derivative with no counterparty risk.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
CVA-DVA
If two counterparties subtracted the CVA of the other to theprice of the default-free portfolio, no deal would be closed(unless one of them was default-free).
This led to the introduction of a Debit ValuationAdjustment (DVA) to take into account the own defaultprobability.
This term raised paradoxical situations: DVAs get larger whenthe own credit quality worsens, increasing the value of ourportfolio since the chances that we will not fully pay the dueamounts are greater.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Example of CVA-DVA
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
CVA-VaR
Changes in CVA-DVA can end up generating serious losses.
Basel III incorporated CVA capital charges based on aValue-at-Risk (VaR) estimate for CVA.
This CVA-VaR is exclusively focused extreme credit spreadoscillations.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Risk factor modeling
Financial entities are typically forced to use conservativeapproaches when calculating credit exposures.
Upon regulatory approval, financial entities can adopt theInternal Model Method (IMM) approach, intended to bettercapture the risk profile of each particular entity.
To do so, they must show their local regulators their ability todescribe the set of risk factors they are exposed to.
One of these factors is the behavior of credit quality.
Thus, realistic and simple credit spread modeling is amandatory milestone for any bank willing to undertake IMMapproval.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
Section 1
Credit spread modeling effects on counterpartyrisk valuation adjustments: a Spanish case study
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
Introduction
Financial entities must accurately model default probabilitiesof all their counterparties with which they have booked a dealin the trading book.
This can easily mean having to compute default probabilitiesfor thousands of different names.
Banks need to find an equilibrium between model complexityand model accuracy.
Moreover, current accountancy standards require the use of asmuch market information as possible when calculating the fairvalue of a derivative.
Therefore, the possibility of calibrating the desired model tomarket instruments must also be taken into account.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
The model
In order to compute accurate counterparty valuationadjustments, we will follow the arbitrage-free valuationframework presented in Brigo et al. (2009).
We will consider a model with stochastic Gaussian interestrates...
Interest rates rt = xt + zt + ϕ(t, α)dxt = −axtdt + σdW1t , x0 = 0dzt = −bztdt + ηdW2t , z0 = 0
(1)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
The model
... and CIR++ default intensities:
Default intensity{λt = yt + ψ(t, β)dyt = κ(µ− yt)dt + ν
√ytdW3t
(2)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
The model
Each participant in the transaction will have its own defaultintensity y it , i ∈ {I ,C}.These intensities will be correlated between them.
Moreover, there will also be correlation between intensitiesand interest rates.
Finally, we will also allow for default time correlation.
If we name τi , i ∈ {I ,C}, the defaulting time, they will beconnected through a Gaussian copula function withcorrelation parameter ρCop.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
Calibrating interest rates
We will first focus on the calibration of interest rates, that willbe performed as in Brigo et al. (2009).
We can calibrate our model to the surface of at-the-money(ATM) swaption volatilities.
The expression for European swaption prices under G2++ isknown. Using this formula, ATM swaption prices can be fittedto the market data surface by minimizing squared errors.
α = argminα
∑i ,j
[SwaptionMktATM(0,Ti , Tj)− SwaptionG2++
ATM (0,Ti , Tj ;α)]2
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
Calibrating default intensities
Default intensities represent a more challenging stage in thecalibration process.
There are no liquid credit instruments to infer impliedvolatilities.
We propose a method to satisfactorily calibrate the dynamicsof default intensities.
We achieved closed form solutions of Credit Default Swaps(CDS) spreads in our framework.
Then, we obtained the variance-covariance matrix of the dailyincrements of these spreads and fitted it to the one observedin the market.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
Calibrating default intensities
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
Calibration window
We intend to capture counterparty valuation adjustments intwo rather different scenarios.
The first date, December 15, 2008, can be regarded as the lastmoment when a Spanish financial institution was consideredto be significantly safer than an average European company.
The second one, June 28, 2012, is placed in the most acutephase of the Spanish recession, when Spain had to accept afinancial bailout from the Eurogroup.
We will observe the credit quality of three references to tracktheir behavior in these two dates: BBVA, Santander andiTraxx Europe.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
Calibration window
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
CVA-DVA calculations
We computed counterparty valuation adjustments for a 10ypar IRS and a 5y CDS contract.
The modeling of the dynamics of credit spreads plays asignificant role in the calculation CVA-DVA when compared tosimpler setups (no credit volatility).
Thus, correctly modeling the behavior of credit quality allowsfinancial entities to price their derivatives accurately.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionThe modelThe calibrationCVA-DVA calculations
CVA-DVA calculations: IRS
2008 2012
iTraxx vs BBVA No Vol Vol No Vol Vol
CVA -0.34274 -0.30463 -1.37362 -1.25786DVA 0.32796 0.43124 0.29015 0.38058
iTraxx vs Santander No Vol Vol No Vol Vol
CVA -0.35579 -0.31230 -1.32088 -1.17327DVA 0.32802 0.43142 0.29021 0.38066
BBVA vs Santander No Vol Vol No Vol Vol
CVA -0.35588 -0.31246 -1.32024 -1.17226DVA 0.18031 0.32321 0.62199 0.74061
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Section 2
Model risk in credit valuation adjustments: aSpanish case study
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Introduction
Regulators are increasingly worrying about sources ofuncertainty around the fair value of a derivative.
The European Banking Authority (EBA) intends that bankscalculate additional valuation adjustments (AVAs) fordetermining the prudent value of fair valued positions.
One of these AVAs pretends to address model risk whencalculating Counterparty Valuation Adjustments (CVA) on thefair value of a derivative.
In this chapter, we will address two sources of model riskunder the framework presented in the previous chapter.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Default correlation
In our modeling framework, we connected defaulting timeswith a Gaussian copula with correlation parameter ρCop.
We had no way to discern which value does ρCop take.
We could also question the use of a Gaussian copula.
However, due to the popularity of Gaussian copulas, we willaddress model risk when correlating defaulting times throughthe correlation parameter ρCop.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Default correlation
We model ρCop with the following distribution:
ρCop =2
1 + eς−ϕ− 1
Due to underlying economic variables, we can assume thatcorrelation in defaulting time will be close to correlation indefault probability (already calibrated).
E [ρCop] =1√2π
∫ +∞
−∞
[ 2
1 + eς−x− 1]e−
12x2dx = ρSpread
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Default correlation
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Default correlation
We found ς such that E [ρCop] = ρSpread .
Once calibrated, we can obtain percentiles for the correlationbetween defaulting times.
We can compute counterparty valuation adjustments withthese percentiles to obtain a confidence interval for CVA-DVA.
Counterparty valuation adjustments are monotone in defaultcorrelation most of the time.
However, there is not a common pattern, either increasing ordecreasing, since they depend also on the survival probabilitiesin the corresponding period.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Default correlation
2007-2008 BBVA-Santander BBVA-iTraxx Santander-iTraxx
ρSpread 91.519% 83.009% 80.445%ς -3.554 -2.761 -2.592
Percentile 5% 74.1889% 50.6593% 44.0956%Percentile 95% 98.9017% 97.5886% 97.1497%
2010-2012 BBVA-Santander BBVA-iTraxx Santander-iTraxx
ρSpread 97.389% 81.982% 81.463%ς -4.804 -2.691 -2.657
Percentile 5% 91.8550% 48.0020% 46.6733%Percentile 95% 99.6841% 97.4154% 97.3265%
Table: Defaulting time correlation parameters. ς has been set such thatthe expected value of default correlation ρCop equals the correlationbetween default intensities ρSpread .
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Default correlation (2008)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Default correlation (2012)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Spread modeling
Default intensities are typically modeled either under a CIR ora Gaussian framework.
In the previous chapter, we used the CIR model to preventexcessively negative default probabilities.
However, we might have used the Gaussian framework due toits higher analytical tractability.
Accountancy rules define fair value as an exit price, so weshould take into account which assumptions may be takingother market players.
Thus, we will alternatively model default intensities with aH&W model.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Spread modeling
Interest rates will be modeled in a G2++ setup, as before.
Default intensities will be described as:λt = yt + ψ(t, β)dyt = −κytdt + νdW3t
y0 = 0
Again, in order to estimate the parameters, we will find aformula for instantaneous CDS spread increments andcalibrate them to those observed in the market.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Spread modeling
When comparing both distributions (CIR and H&W), weobserve that they yield similar masses of negative defaultprobabilities.
However, the H&W model produces more extreme values ofnegative probabilities.
These differences end up translating into variations ofcounterparty valuation adjustments.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Survival probabilities (2012)
H&W++ CIR++
PS(5y , 6y) BBVA Santander iTraxx BBVA Santander iTraxx
Mean 0.9012 0.9072 0.9618 0.9019 0.9076 0.9619Std 0.0725 0.0708 0.0255 0.0722 0.0709 0.0250
Ex. kurtosis 0.1200 0.1065 0.0380 0.0588 0.1777 0.6220Skewness 0.2384 0.2354 0.0721 -0.3972 -0.4893 -0.7114
Min 0.6328 0.6490 0.8486 0.5679 0.5197 0.8106Max 1.2698 1.2497 1.0802 1.0931 1.0754 1.0137
Percentile 1% 0.7458 0.7548 0.9036 0.7166 0.7215 0.8917Percentile 5% 0.7865 0.7958 0.9206 0.7747 0.7808 0.9160Percentile 95% 1.0247 1.0277 1.0042 1.0113 1.0132 0.9971Percentile 99% 1.0833 1.0841 1.0225 1.0440 1.0422 1.0055Q{PS < 1} 0.9097 0.9010 0.9305 0.9254 0.9166 0.9684
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
IntroductionDefault correlationSpread modeling
Survival probabilities (2012)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Section 3
Credit spread modeling from a regulatoryperspective
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Calculating VaR for CVA
Basel III presents a metric to address potential CVA-relatedlosses.
Based upon a given formula, Basel III proposes the calculationof a CVA-VaR.
Although CVA is dependent on LGD and expected exposures,VaR shall rely exclusively on credit spreads.
Basel III recipe for CVA
CVA = LGD ·∑T
i=1 max(
0; exp(− si−1ti−1
LGD
)−
exp(− si ti
LGD
))·(EEi−1Di−1+EEiDi
2
)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Calculating VaR for CVA
We calculated a parametric and a historical CVA-VaR alongtwo years (2012-2013).
We calibrate the model for interest rates and defaultintensities on a monthly basis.
Three references are studied: BBVA, Santander and iTraxxEurope.
Expected positive exposure profile is that of a par 10-yearswap, held constant along the year.
We will test the VaR model along a one-year horizon.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
A parametric VaR
Get the interest rate curve and CDS spreads for day i .
Get the corresponding G2++ and CIR++ model parameters.
Simulate 10,000 times one day of interest rates and defaultintensities and obtain a distribution of CDS spreads for dayi + 1.
From these possible CDS spreads, calculate a distribution ofpossible CVA values.
Take percentile 99% of these CVA values as the CVA VaR.
Take the realized CDS spreads of day i + 1 and compute CVAfrom these spreads. If it is greater than the previouslycalculated VaR, it will be considered an exception.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Calculating VaR for CVA
The historical VaR appears to be conservative when comparedto the parametric one.
The parametric VaR scores yellow for BBVA and Santander in2012.
That year was specially troubled for Spanish financialinstitutions.
The historical VaR was also close to fall in the yellow zone in2012 for these references, but narrowly missed it.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Calculating VaR for CVA
2012 Historical Parametric
Entity Nb Exceptions Zone Nb Exceptions Zone
BBVA 4 Green 6 YellowSantander 2 Green 5 Yellow
iTraxx 0 Green 0 Green
2013 Historical Parametric
Entity Nb Exceptions Zone Nb Exceptions Zone
BBVA 2 Green 3 GreenSantander 2 Green 3 Green
iTraxx 1 Green 2 Green
Table: Number of exceptions for CVA VaR (2012-2013)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Calculating VaR for CVA (BBVA)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Introduction
Financial entities have an incentive to migrate to InternalModel Method (IMM) in order to optimize their capitalrequirements.
To do so, banks must show their regulators the suitability oftheir models to describe the set of risk factors they areexposed to.
Backtesting a risk factor means comparing the realizedhistorical distributions with the ones generated via simulationunder the proposed model.
We need to focus on the whole exposure distribution, ratherthan the tails as in a VaR approach.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
A first approximation to backtesting
We define:
A historical window T = tend − tstartA recalibration frequency t = [tstart , t1 = t0 + ∆, ..., tend ]A time horizon ∆ over which study our model.
We then proceed as follows:
Go to recalibration node ti . Calculate the model risk factordistribution at point ti + ∆.Observe the realized value of the risk factor at ti + ∆ andcalculate its probability Fi .Repeat the above until ti + ∆ reaches tend .
Remark
If reality behaved as our model, the collection {Fi}Ni=1 wouldbe uniformly distributed
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
A first approximation to backtesting
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
A first approximation to backtesting
We define a distance function D between distributions tocheck how far {Fi}Ni=1 is from a uniform distribution.
Now, Basel III recommends to backtest horizons of up to oneyear.
In a two-year sample, this would leave us with only twoobservations.
We need to extend the backtesting methodology followingKenyon (2012) and Ruiz (2012).
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Extending the backtesting method
We define:
A historical window T = tend − tstartA recalibration frequency t = [tstart , t1 = t0 + δ, ..., tend ]A time horizon ∆ over which study our model, with δ ≤ ∆.
As before, we generate a collection of probabilities {Fi}Ni=1,that will not be uniformly distributed due to auto-correlationeffects:
Eg: δ = 1 month and ∆ = 1 year along 2012 and 2013.On January 1, 2012 we estimate for January 1, 2013, observethe realized value on the latter date and get a probability of40%.On the following recalibration date, February 1, 2012, weestimate for February 1, 2013.The estimated probability will be again close to 40%.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Extending the backtesting method
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Extending the backtesting method
For very large historical windows, this effect would eventuallyvanish, but not in finite samples (as it is our case).
We will create M pseudo-historical paths with our model andcompute the corresponding distances to a uniform distribution{Dk}Mk=1, compatible with a perfect model.
This collection will follow a distribution Υ(D).
This allows us to assign a probability to the historical distance.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Results
We performed a backtesting exercise along 2012-2013 withmonthly calibrations for the CIR model.
Applying the first methodology to 1m horizon, we did notreject the CIR distribution for the CDS 5y.
Applying the extended methodology to larger horizons, we didnot reject the CIR distribution.
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Results
Figure: 5%-95% confidence intervals under the model distribution forCDS5y vs realized value. BBVA. ∆ = 1 month
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
Credit spread modeling effects on CVA: a Spanish case studyModel risk in CVA: a Spanish case study
Credit spread modeling from a regulatory perspective
Calculating VaR for CVABacktesting a risk factor under IMM
Backtesting results for BBVA (2012-2013)
Alberto Fernandez Munoz de Morales Three essays on the cutting edge of credit spread modeling
top related